ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/slab.h>
  20 #include <linux/mm.h>
  21 #include <linux/hugetlb.h>
  22 #include <linux/shm.h>
  23 #include <linux/init.h>
  24 #include <linux/file.h>
  25 #include <linux/mman.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/shmem_fs.h>
  28 #include <linux/security.h>
  29 #include <asm/uaccess.h>
  30
  31 #include "util.h"
  32
  33 #define shm_flags       shm_perm.mode
  34
  35 static struct file_operations shm_file_operations;
  36 static struct vm_operations_struct shm_vm_ops;
  37
  38 static struct ipc_ids shm_ids;
  39
  40 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  41 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  42 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  43 #define shm_buildid(id, seq) \
  44         ipc_buildid(&shm_ids, id, seq)
  45
  46 static int newseg (key_t key, int shmflg, size_t size);
  47 static void shm_open (struct vm_area_struct *shmd);
  48 static void shm_close (struct vm_area_struct *shmd);
  49 #ifdef CONFIG_PROC_FS
  50 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  51 #endif
  52
  53 size_t  shm_ctlmax = SHMMAX;
  54 size_t  shm_ctlall = SHMALL;
  55 int     shm_ctlmni = SHMMNI;
  56
  57 static int shm_tot; /* total number of shared memory pages */
  58
  59 void __init shm_init (void)
  60 {
  61         ipc_init_ids(&shm_ids, 1);
  62 #ifdef CONFIG_PROC_FS
  63         create_proc_read_entry("sysvipc/shm", 0, NULL, sysvipc_shm_read_proc, NULL);
  64 #endif
  65 }
  66
  67 static inline int shm_checkid(struct shmid_kernel *s, int id)
  68 {
  69         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  70                 return -EIDRM;
  71         return 0;
  72 }
  73
  74 static inline struct shmid_kernel *shm_rmid(int id)
  75 {
  76         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  77 }
  78
  79 static inline int shm_addid(struct shmid_kernel *shp)
  80 {
  81         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  82 }
  83
  84
  85
  86 static inline void shm_inc (int id) {
  87         struct shmid_kernel *shp;
  88
  89         if(!(shp = shm_lock(id)))
  90                 BUG();
  91         shp->shm_atim = get_seconds();
  92         shp->shm_lprid = current->tgid;
  93         shp->shm_nattch++;
  94         shm_unlock(shp);
  95 }
  96
  97 /* This is called by fork, once for every shm attach. */
  98 static void shm_open (struct vm_area_struct *shmd)
  99 {
 100         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 101 }
 102
 103 /*
 104  * shm_destroy - free the struct shmid_kernel
 105  *
 106  * @shp: struct to free
 107  *
 108  * It has to be called with shp and shm_ids.sem locked,
 109  * but returns with shp unlocked and freed.
 110  */
 111 static void shm_destroy (struct shmid_kernel *shp)
 112 {
 113         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 114         shm_rmid (shp->id);
 115         shm_unlock(shp);
 116         if (!is_file_hugepages(shp->shm_file))
 117                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 118         else
 119                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 120                                                 shp->mlock_user);
 121         fput (shp->shm_file);
 122         security_shm_free(shp);
 123         ipc_rcu_putref(shp);
 124 }
 125
 126 /*
 127  * remove the attach descriptor shmd.
 128  * free memory for segment if it is marked destroyed.
 129  * The descriptor has already been removed from the current->mm->mmap list
 130  * and will later be kfree()d.
 131  */
 132 static void shm_close (struct vm_area_struct *shmd)
 133 {
 134         struct file * file = shmd->vm_file;
 135         int id = file->f_dentry->d_inode->i_ino;
 136         struct shmid_kernel *shp;
 137
 138         down (&shm_ids.sem);
 139         /* remove from the list of attaches of the shm segment */
 140         if(!(shp = shm_lock(id)))
 141                 BUG();
 142         shp->shm_lprid = current->tgid;
 143         shp->shm_dtim = get_seconds();
 144         shp->shm_nattch--;
 145         if(shp->shm_nattch == 0 &&
 146            shp->shm_flags & SHM_DEST)
 147                 shm_destroy (shp);
 148         else
 149                 shm_unlock(shp);
 150         up (&shm_ids.sem);
 151 }
 152
 153 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 154 {
 155         file_accessed(file);
 156         vma->vm_ops = &shm_vm_ops;
 157         shm_inc(file->f_dentry->d_inode->i_ino);
 158         return 0;
 159 }
 160
 161 static struct file_operations shm_file_operations = {
 162         .mmap   = shm_mmap
 163 };
 164
 165 static struct vm_operations_struct shm_vm_ops = {
 166         .open   = shm_open,     /* callback for a new vm-area open */
 167         .close  = shm_close,    /* callback for when the vm-area is released */
 168         .nopage = shmem_nopage,
 169 #ifdef CONFIG_NUMA
 170         .set_policy = shmem_set_policy,
 171         .get_policy = shmem_get_policy,
 172 #endif
 173 };
 174
 175 static int newseg (key_t key, int shmflg, size_t size)
 176 {
 177         int error;
 178         struct shmid_kernel *shp;
 179         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 180         struct file * file;
 181         char name[13];
 182         int id;
 183
 184         if (size < SHMMIN || size > shm_ctlmax)
 185                 return -EINVAL;
 186
 187         if (shm_tot + numpages >= shm_ctlall)
 188                 return -ENOSPC;
 189
 190         shp = ipc_rcu_alloc(sizeof(*shp));
 191         if (!shp)
 192                 return -ENOMEM;
 193
 194         shp->shm_perm.key = key;
 195         shp->shm_flags = (shmflg & S_IRWXUGO);
 196         shp->mlock_user = NULL;
 197
 198         shp->shm_perm.security = NULL;
 199         error = security_shm_alloc(shp);
 200         if (error) {
 201                 ipc_rcu_putref(shp);
 202                 return error;
 203         }
 204
 205         if (shmflg & SHM_HUGETLB) {
 206                 /* hugetlb_zero_setup takes care of mlock user accounting */
 207                 file = hugetlb_zero_setup(size);
 208                 shp->mlock_user = current->user;
 209         } else {
 210                 sprintf (name, "SYSV%08x", key);
 211                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 212         }
 213         error = PTR_ERR(file);
 214         if (IS_ERR(file))
 215                 goto no_file;
 216
 217         error = -ENOSPC;
 218         id = shm_addid(shp);
 219         if(id == -1)
 220                 goto no_id;
 221
 222         shp->shm_cprid = current->tgid;
 223         shp->shm_lprid = 0;
 224         shp->shm_atim = shp->shm_dtim = 0;
 225         shp->shm_ctim = get_seconds();
 226         shp->shm_segsz = size;
 227         shp->shm_nattch = 0;
 228         shp->id = shm_buildid(id,shp->shm_perm.seq);
 229         shp->shm_file = file;
 230         file->f_dentry->d_inode->i_ino = shp->id;
 231         if (shmflg & SHM_HUGETLB)
 232                 set_file_hugepages(file);
 233         else
 234                 file->f_op = &shm_file_operations;
 235         shm_tot += numpages;
 236         shm_unlock(shp);
 237         return shp->id;
 238
 239 no_id:
 240         fput(file);
 241 no_file:
 242         security_shm_free(shp);
 243         ipc_rcu_putref(shp);
 244         return error;
 245 }
 246
 247 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 248 {
 249         struct shmid_kernel *shp;
 250         int err, id = 0;
 251
 252         down(&shm_ids.sem);
 253         if (key == IPC_PRIVATE) {
 254                 err = newseg(key, shmflg, size);
 255         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 256                 if (!(shmflg & IPC_CREAT))
 257                         err = -ENOENT;
 258                 else
 259                         err = newseg(key, shmflg, size);
 260         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 261                 err = -EEXIST;
 262         } else {
 263                 shp = shm_lock(id);
 264                 if(shp==NULL)
 265                         BUG();
 266                 if (shp->shm_segsz < size)
 267                         err = -EINVAL;
 268                 else if (ipcperms(&shp->shm_perm, shmflg))
 269                         err = -EACCES;
 270                 else {
 271                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 272                         err = security_shm_associate(shp, shmflg);
 273                         if (!err)
 274                                 err = shmid;
 275                 }
 276                 shm_unlock(shp);
 277         }
 278         up(&shm_ids.sem);
 279
 280         return err;
 281 }
 282
 283 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 284 {
 285         switch(version) {
 286         case IPC_64:
 287                 return copy_to_user(buf, in, sizeof(*in));
 288         case IPC_OLD:
 289             {
 290                 struct shmid_ds out;
 291
 292                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 293                 out.shm_segsz   = in->shm_segsz;
 294                 out.shm_atime   = in->shm_atime;
 295                 out.shm_dtime   = in->shm_dtime;
 296                 out.shm_ctime   = in->shm_ctime;
 297                 out.shm_cpid    = in->shm_cpid;
 298                 out.shm_lpid    = in->shm_lpid;
 299                 out.shm_nattch  = in->shm_nattch;
 300
 301                 return copy_to_user(buf, &out, sizeof(out));
 302             }
 303         default:
 304                 return -EINVAL;
 305         }
 306 }
 307
 308 struct shm_setbuf {
 309         uid_t   uid;
 310         gid_t   gid;
 311         mode_t  mode;
 312 };
 313
 314 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 315 {
 316         switch(version) {
 317         case IPC_64:
 318             {
 319                 struct shmid64_ds tbuf;
 320
 321                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 322                         return -EFAULT;
 323
 324                 out->uid        = tbuf.shm_perm.uid;
 325                 out->gid        = tbuf.shm_perm.gid;
 326                 out->mode       = tbuf.shm_flags;
 327
 328                 return 0;
 329             }
 330         case IPC_OLD:
 331             {
 332                 struct shmid_ds tbuf_old;
 333
 334                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 335                         return -EFAULT;
 336
 337                 out->uid        = tbuf_old.shm_perm.uid;
 338                 out->gid        = tbuf_old.shm_perm.gid;
 339                 out->mode       = tbuf_old.shm_flags;
 340
 341                 return 0;
 342             }
 343         default:
 344                 return -EINVAL;
 345         }
 346 }
 347
 348 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 349 {
 350         switch(version) {
 351         case IPC_64:
 352                 return copy_to_user(buf, in, sizeof(*in));
 353         case IPC_OLD:
 354             {
 355                 struct shminfo out;
 356
 357                 if(in->shmmax > INT_MAX)
 358                         out.shmmax = INT_MAX;
 359                 else
 360                         out.shmmax = (int)in->shmmax;
 361
 362                 out.shmmin      = in->shmmin;
 363                 out.shmmni      = in->shmmni;
 364                 out.shmseg      = in->shmseg;
 365                 out.shmall      = in->shmall;
 366
 367                 return copy_to_user(buf, &out, sizeof(out));
 368             }
 369         default:
 370                 return -EINVAL;
 371         }
 372 }
 373
 374 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 375 {
 376         int i;
 377
 378         *rss = 0;
 379         *swp = 0;
 380
 381         for (i = 0; i <= shm_ids.max_id; i++) {
 382                 struct shmid_kernel *shp;
 383                 struct inode *inode;
 384
 385                 shp = shm_get(i);
 386                 if(!shp)
 387                         continue;
 388
 389                 inode = shp->shm_file->f_dentry->d_inode;
 390
 391                 if (is_file_hugepages(shp->shm_file)) {
 392                         struct address_space *mapping = inode->i_mapping;
 393                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 394                 } else {
 395                         struct shmem_inode_info *info = SHMEM_I(inode);
 396                         spin_lock(&info->lock);
 397                         *rss += inode->i_mapping->nrpages;
 398                         *swp += info->swapped;
 399                         spin_unlock(&info->lock);
 400                 }
 401         }
 402 }
 403
 404 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 405 {
 406         struct shm_setbuf setbuf;
 407         struct shmid_kernel *shp;
 408         int err, version;
 409
 410         if (cmd < 0 || shmid < 0) {
 411                 err = -EINVAL;
 412                 goto out;
 413         }
 414
 415         version = ipc_parse_version(&cmd);
 416
 417         switch (cmd) { /* replace with proc interface ? */
 418         case IPC_INFO:
 419         {
 420                 struct shminfo64 shminfo;
 421
 422                 err = security_shm_shmctl(NULL, cmd);
 423                 if (err)
 424                         return err;
 425
 426                 memset(&shminfo,0,sizeof(shminfo));
 427                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 428                 shminfo.shmmax = shm_ctlmax;
 429                 shminfo.shmall = shm_ctlall;
 430
 431                 shminfo.shmmin = SHMMIN;
 432                 if(copy_shminfo_to_user (buf, &shminfo, version))
 433                         return -EFAULT;
 434                 /* reading a integer is always atomic */
 435                 err= shm_ids.max_id;
 436                 if(err<0)
 437                         err = 0;
 438                 goto out;
 439         }
 440         case SHM_INFO:
 441         {
 442                 struct shm_info shm_info;
 443
 444                 err = security_shm_shmctl(NULL, cmd);
 445                 if (err)
 446                         return err;
 447
 448                 memset(&shm_info,0,sizeof(shm_info));
 449                 down(&shm_ids.sem);
 450                 shm_info.used_ids = shm_ids.in_use;
 451                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 452                 shm_info.shm_tot = shm_tot;
 453                 shm_info.swap_attempts = 0;
 454                 shm_info.swap_successes = 0;
 455                 err = shm_ids.max_id;
 456                 up(&shm_ids.sem);
 457                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 458                         err = -EFAULT;
 459                         goto out;
 460                 }
 461
 462                 err = err < 0 ? 0 : err;
 463                 goto out;
 464         }
 465         case SHM_STAT:
 466         case IPC_STAT:
 467         {
 468                 struct shmid64_ds tbuf;
 469                 int result;
 470                 memset(&tbuf, 0, sizeof(tbuf));
 471                 shp = shm_lock(shmid);
 472                 if(shp==NULL) {
 473                         err = -EINVAL;
 474                         goto out;
 475                 } else if(cmd==SHM_STAT) {
 476                         err = -EINVAL;
 477                         if (shmid > shm_ids.max_id)
 478                                 goto out_unlock;
 479                         result = shm_buildid(shmid, shp->shm_perm.seq);
 480                 } else {
 481                         err = shm_checkid(shp,shmid);
 482                         if(err)
 483                                 goto out_unlock;
 484                         result = 0;
 485                 }
 486                 err=-EACCES;
 487                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 488                         goto out_unlock;
 489                 err = security_shm_shmctl(shp, cmd);
 490                 if (err)
 491                         goto out_unlock;
 492                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 493                 tbuf.shm_segsz  = shp->shm_segsz;
 494                 tbuf.shm_atime  = shp->shm_atim;
 495                 tbuf.shm_dtime  = shp->shm_dtim;
 496                 tbuf.shm_ctime  = shp->shm_ctim;
 497                 tbuf.shm_cpid   = shp->shm_cprid;
 498                 tbuf.shm_lpid   = shp->shm_lprid;
 499                 if (!is_file_hugepages(shp->shm_file))
 500                         tbuf.shm_nattch = shp->shm_nattch;
 501                 else
 502                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 503                 shm_unlock(shp);
 504                 if(copy_shmid_to_user (buf, &tbuf, version))
 505                         err = -EFAULT;
 506                 else
 507                         err = result;
 508                 goto out;
 509         }
 510         case SHM_LOCK:
 511         case SHM_UNLOCK:
 512         {
 513                 /* Allow superuser to lock segment in memory */
 514                 if (!can_do_mlock() && cmd == SHM_LOCK) {
 515                         err = -EPERM;
 516                         goto out;
 517                 }
 518                 shp = shm_lock(shmid);
 519                 if(shp==NULL) {
 520                         err = -EINVAL;
 521                         goto out;
 522                 }
 523                 err = shm_checkid(shp,shmid);
 524                 if(err)
 525                         goto out_unlock;
 526
 527                 err = security_shm_shmctl(shp, cmd);
 528                 if (err)
 529                         goto out_unlock;
 530
 531                 if(cmd==SHM_LOCK) {
 532                         struct user_struct * user = current->user;
 533                         if (!is_file_hugepages(shp->shm_file)) {
 534                                 err = shmem_lock(shp->shm_file, 1, user);
 535                                 if (!err) {
 536                                         shp->shm_flags |= SHM_LOCKED;
 537                                         shp->mlock_user = user;
 538                                 }
 539                         }
 540                 } else if (!is_file_hugepages(shp->shm_file)) {
 541                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 542                         shp->shm_flags &= ~SHM_LOCKED;
 543                         shp->mlock_user = NULL;
 544                 }
 545                 shm_unlock(shp);
 546                 goto out;
 547         }
 548         case IPC_RMID:
 549         {
 550                 /*
 551                  *      We cannot simply remove the file. The SVID states
 552                  *      that the block remains until the last person
 553                  *      detaches from it, then is deleted. A shmat() on
 554                  *      an RMID segment is legal in older Linux and if
 555                  *      we change it apps break...
 556                  *
 557                  *      Instead we set a destroyed flag, and then blow
 558                  *      the name away when the usage hits zero.
 559                  */
 560                 down(&shm_ids.sem);
 561                 shp = shm_lock(shmid);
 562                 err = -EINVAL;
 563                 if (shp == NULL)
 564                         goto out_up;
 565                 err = shm_checkid(shp, shmid);
 566                 if(err)
 567                         goto out_unlock_up;
 568
 569                 if (current->euid != shp->shm_perm.uid &&
 570                     current->euid != shp->shm_perm.cuid &&
 571                     !capable(CAP_SYS_ADMIN)) {
 572                         err=-EPERM;
 573                         goto out_unlock_up;
 574                 }
 575
 576                 err = security_shm_shmctl(shp, cmd);
 577                 if (err)
 578                         goto out_unlock_up;
 579
 580                 if (shp->shm_nattch){
 581                         shp->shm_flags |= SHM_DEST;
 582                         /* Do not find it any more */
 583                         shp->shm_perm.key = IPC_PRIVATE;
 584                         shm_unlock(shp);
 585                 } else
 586                         shm_destroy (shp);
 587                 up(&shm_ids.sem);
 588                 goto out;
 589         }
 590
 591         case IPC_SET:
 592         {
 593                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 594                         err = -EFAULT;
 595                         goto out;
 596                 }
 597                 down(&shm_ids.sem);
 598                 shp = shm_lock(shmid);
 599                 err=-EINVAL;
 600                 if(shp==NULL)
 601                         goto out_up;
 602                 err = shm_checkid(shp,shmid);
 603                 if(err)
 604                         goto out_unlock_up;
 605                 err=-EPERM;
 606                 if (current->euid != shp->shm_perm.uid &&
 607                     current->euid != shp->shm_perm.cuid &&
 608                     !capable(CAP_SYS_ADMIN)) {
 609                         goto out_unlock_up;
 610                 }
 611
 612                 err = security_shm_shmctl(shp, cmd);
 613                 if (err)
 614                         goto out_unlock_up;
 615
 616                 shp->shm_perm.uid = setbuf.uid;
 617                 shp->shm_perm.gid = setbuf.gid;
 618                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 619                         | (setbuf.mode & S_IRWXUGO);
 620                 shp->shm_ctim = get_seconds();
 621                 break;
 622         }
 623
 624         default:
 625                 err = -EINVAL;
 626                 goto out;
 627         }
 628
 629         err = 0;
 630 out_unlock_up:
 631         shm_unlock(shp);
 632 out_up:
 633         up(&shm_ids.sem);
 634         goto out;
 635 out_unlock:
 636         shm_unlock(shp);
 637 out:
 638         return err;
 639 }
 640
 641 /*
 642  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 643  *
 644  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 645  * "raddr" thing points to kernel space, and there has to be a wrapper around
 646  * this.
 647  */
 648 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 649 {
 650         struct shmid_kernel *shp;
 651         unsigned long addr;
 652         unsigned long size;
 653         struct file * file;
 654         int    err;
 655         unsigned long flags;
 656         unsigned long prot;
 657         unsigned long o_flags;
 658         int acc_mode;
 659         void *user_addr;
 660
 661         if (shmid < 0) {
 662                 err = -EINVAL;
 663                 goto out;
 664         } else if ((addr = (ulong)shmaddr)) {
 665                 if (addr & (SHMLBA-1)) {
 666                         if (shmflg & SHM_RND)
 667                                 addr &= ~(SHMLBA-1);       /* round down */
 668                         else
 669 #ifndef __ARCH_FORCE_SHMLBA
 670                                 if (addr & ~PAGE_MASK)
 671 #endif
 672                                         return -EINVAL;
 673                 }
 674                 flags = MAP_SHARED | MAP_FIXED;
 675         } else {
 676                 if ((shmflg & SHM_REMAP))
 677                         return -EINVAL;
 678
 679                 flags = MAP_SHARED;
 680         }
 681
 682         if (shmflg & SHM_RDONLY) {
 683                 prot = PROT_READ;
 684                 o_flags = O_RDONLY;
 685                 acc_mode = S_IRUGO;
 686         } else {
 687                 prot = PROT_READ | PROT_WRITE;
 688                 o_flags = O_RDWR;
 689                 acc_mode = S_IRUGO | S_IWUGO;
 690         }
 691         if (shmflg & SHM_EXEC) {
 692                 prot |= PROT_EXEC;
 693                 acc_mode |= S_IXUGO;
 694         }
 695
 696         /*
 697          * We cannot rely on the fs check since SYSV IPC does have an
 698          * additional creator id...
 699          */
 700         shp = shm_lock(shmid);
 701         if(shp == NULL) {
 702                 err = -EINVAL;
 703                 goto out;
 704         }
 705         err = shm_checkid(shp,shmid);
 706         if (err) {
 707                 shm_unlock(shp);
 708                 goto out;
 709         }
 710         if (ipcperms(&shp->shm_perm, acc_mode)) {
 711                 shm_unlock(shp);
 712                 err = -EACCES;
 713                 goto out;
 714         }
 715
 716         err = security_shm_shmat(shp, shmaddr, shmflg);
 717         if (err) {
 718                 shm_unlock(shp);
 719                 return err;
 720         }
 721
 722         file = shp->shm_file;
 723         size = i_size_read(file->f_dentry->d_inode);
 724         shp->shm_nattch++;
 725         shm_unlock(shp);
 726
 727         down_write(&current->mm->mmap_sem);
 728         if (addr && !(shmflg & SHM_REMAP)) {
 729                 user_addr = ERR_PTR(-EINVAL);
 730                 if (find_vma_intersection(current->mm, addr, addr + size))
 731                         goto invalid;
 732                 /*
 733                  * If shm segment goes below stack, make sure there is some
 734                  * space left for the stack to grow (at least 4 pages).
 735                  */
 736                 if (addr < current->mm->start_stack &&
 737                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 738                         goto invalid;
 739         }
 740
 741         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 742
 743 invalid:
 744         up_write(&current->mm->mmap_sem);
 745
 746         down (&shm_ids.sem);
 747         if(!(shp = shm_lock(shmid)))
 748                 BUG();
 749         shp->shm_nattch--;
 750         if(shp->shm_nattch == 0 &&
 751            shp->shm_flags & SHM_DEST)
 752                 shm_destroy (shp);
 753         else
 754                 shm_unlock(shp);
 755         up (&shm_ids.sem);
 756
 757         *raddr = (unsigned long) user_addr;
 758         err = 0;
 759         if (IS_ERR(user_addr))
 760                 err = PTR_ERR(user_addr);
 761 out:
 762         return err;
 763 }
 764
 765 /*
 766  * detach and kill segment if marked destroyed.
 767  * The work is done in shm_close.
 768  */
 769 asmlinkage long sys_shmdt(char __user *shmaddr)
 770 {
 771         struct mm_struct *mm = current->mm;
 772         struct vm_area_struct *vma, *next;
 773         unsigned long addr = (unsigned long)shmaddr;
 774         loff_t size = 0;
 775         int retval = -EINVAL;
 776
 777         down_write(&mm->mmap_sem);
 778
 779         /*
 780          * This function tries to be smart and unmap shm segments that
 781          * were modified by partial mlock or munmap calls:
 782          * - It first determines the size of the shm segment that should be
 783          *   unmapped: It searches for a vma that is backed by shm and that
 784          *   started at address shmaddr. It records it's size and then unmaps
 785          *   it.
 786          * - Then it unmaps all shm vmas that started at shmaddr and that
 787          *   are within the initially determined size.
 788          * Errors from do_munmap are ignored: the function only fails if
 789          * it's called with invalid parameters or if it's called to unmap
 790          * a part of a vma. Both calls in this function are for full vmas,
 791          * the parameters are directly copied from the vma itself and always
 792          * valid - therefore do_munmap cannot fail. (famous last words?)
 793          */
 794         /*
 795          * If it had been mremap()'d, the starting address would not
 796          * match the usual checks anyway. So assume all vma's are
 797          * above the starting address given.
 798          */
 799         vma = find_vma(mm, addr);
 800
 801         while (vma) {
 802                 next = vma->vm_next;
 803
 804                 /*
 805                  * Check if the starting address would match, i.e. it's
 806                  * a fragment created by mprotect() and/or munmap(), or it
 807                  * otherwise it starts at this address with no hassles.
 808                  */
 809                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 810                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 811
 812
 813                         size = vma->vm_file->f_dentry->d_inode->i_size;
 814                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 815                         /*
 816                          * We discovered the size of the shm segment, so
 817                          * break out of here and fall through to the next
 818                          * loop that uses the size information to stop
 819                          * searching for matching vma's.
 820                          */
 821                         retval = 0;
 822                         vma = next;
 823                         break;
 824                 }
 825                 vma = next;
 826         }
 827
 828         /*
 829          * We need look no further than the maximum address a fragment
 830          * could possibly have landed at. Also cast things to loff_t to
 831          * prevent overflows and make comparisions vs. equal-width types.
 832          */
 833         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 834                 next = vma->vm_next;
 835
 836                 /* finding a matching vma now does not alter retval */
 837                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 838                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 839
 840                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 841                 vma = next;
 842         }
 843
 844         up_write(&mm->mmap_sem);
 845         return retval;
 846 }
 847
 848 #ifdef CONFIG_PROC_FS
 849 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 850 {
 851         off_t pos = 0;
 852         off_t begin = 0;
 853         int i, len = 0;
 854
 855         down(&shm_ids.sem);
 856         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 857
 858         for(i = 0; i <= shm_ids.max_id; i++) {
 859                 struct shmid_kernel* shp;
 860
 861                 shp = shm_lock(i);
 862                 if(shp!=NULL) {
 863 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 864 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 865                         char *format;
 866
 867                         if (sizeof(size_t) <= sizeof(int))
 868                                 format = SMALL_STRING;
 869                         else
 870                                 format = BIG_STRING;
 871                         len += sprintf(buffer + len, format,
 872                                 shp->shm_perm.key,
 873                                 shm_buildid(i, shp->shm_perm.seq),
 874                                 shp->shm_flags,
 875                                 shp->shm_segsz,
 876                                 shp->shm_cprid,
 877                                 shp->shm_lprid,
 878                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 879                                 shp->shm_perm.uid,
 880                                 shp->shm_perm.gid,
 881                                 shp->shm_perm.cuid,
 882                                 shp->shm_perm.cgid,
 883                                 shp->shm_atim,
 884                                 shp->shm_dtim,
 885                                 shp->shm_ctim);
 886                         shm_unlock(shp);
 887
 888                         pos += len;
 889                         if(pos < offset) {
 890                                 len = 0;
 891                                 begin = pos;
 892                         }
 893                         if(pos > offset + length)
 894                                 goto done;
 895                 }
 896         }
 897         *eof = 1;
 898 done:
 899         up(&shm_ids.sem);
 900         *start = buffer + (offset - begin);
 901         len -= (offset - begin);
 902         if(len > length)
 903                 len = length;
 904         if(len < 0)
 905                 len = 0;
 906         return len;
 907 }
 908 #endif